Session: 06-12-03: Robotics, Rehabilitation

Paper Number: 114072

114072 - Robot-Based Adaptive Training of a Repetitive Motion Shows the Potential to Outperform Transient, Passive and Active Learning 

This continuing research studies the relationship between humans and robots from the perspective of autonomy and presents a novel concept of teaching a human a new, repetitive motion (i.e., develop their muscle memory). The research moves beyond prior published work that uses a dynamically autonomous robot (Type X robot) to train a human via transient training, a progressive type of robotic-assisted training where the assistance of the robot is strong at first but fades away over time. This work raises a new concept, adaptive training, where the compensation of the robotic assistance is multi-factor based and real-time evaluated.

 

In the authors’ previous published work, a dynamically autonomous Type X robot has been built and tested. The same robot is further improved and implemented in this work. The robot consists of two levers, each connected to a DC torque motor. The desired motion involves having the subject need move the two levers with sinusoidal motion of amplitude 30 degrees and with the right lever at twice the frequency of the left lever. A TV and speaker system provides the visual (the subject can see the trace created by the lever motion, where left lever controls the cursor’s horizontal position on the screen, and the right lever controls the cursor’s horizontal motion on the screen) and auditory feedback signals (essentially a metronome to establish the desired frequency) needed for the subject to trace a desired Lissajous curve (resembling a figure “8”).

 

The Type X robot was tested by randomly assigning 16 subjects into four different groups: active learning (zero robot assist), passive learning (complete robot assist), transient learning (dynamic robot assist progressing from complete at the start of training towards no assist at the completion of training), and adaptive learning (proposed in this work).  Each subject underwent 90 minutes of training per day (broken into six segments of 15 minutes each) on each of five different days.  At the end of each 15-minute training segment, an assessment test (with no assist from the robot) was conducted to assess how well the desired motion was being learned.  The motion data during each assessment was collected to allow comparison of the various subjects’ performances on drawing the desired Lissajous curve over the duration of their 450 minutes of training.  A series of metrics were developed to enable comprehensive comparison of the subjects’ performances. MATLAB and Minitab are the main data analyzing tools.

 

In this work, one additional subject was tested using the adaptive training approach, wherein the trainer (the first author) closely observed the subjects’ performance on each assessment test and compute what level of autonomy to use during the next training session based on developed training algorithms.

 

The initial result of the research demonstrates that subject in adaptive learning group demonstrated clear and significant performance increases in comparison to all other groups (transient, active and passive learning). In specific, the outcome data suggests that subject in adaptive learning group performs the best for the bimanual repetitive motion in terms of accuracy, synchronization, precision, proficiency of muscle memory, and accumulative error.

 

The implications of this research are potentially profound and far-reaching, with applications including rehabilitation, sports training, etc.

Presenting Author: Jonathan Weaver University of Detroit Mercy

Presenting Author Biography: Professor of Mechanical Engineering Department, University of Detroit Mercy

Authors:

Danqing Zhang University of Detroit Mercy
Jonathan Weaver University of Detroit Mercy